1// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py 2// RUN: mlir-opt %s -sparsification | FileCheck %s 3 4#Td = #sparse_tensor.encoding<{ dimLevelType = [ "dense" ] }> 5 6#Tddd = #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "dense" ] }> 7#Tdds = #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ] }> 8#Tdsd = #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ] }> 9#Tdss = #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ] }> 10#Tsdd = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ] }> 11#Tsds = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ] }> 12#Tssd = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ] }> 13#Tsss = #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ] }> 14 15#trait3 = { 16 indexing_maps = [ 17 affine_map<(i,j,k) -> (i,j,k)>, // A 18 affine_map<(i,j,k) -> (i,j,k)>, // B 19 affine_map<(i,j,k) -> (i,j,k)> // X (out) 20 ], 21 iterator_types = ["parallel", "parallel", "parallel"], 22 doc = "X(i,j,k) = A(i,j,k) OP B(i,j,k)" 23} 24 25// CHECK-LABEL: func @add_ddd( 26// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 27// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 28// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 29// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 30// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 31// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 32// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 33// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 34// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 35// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 36// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 37// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 38// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_11]] : memref<32x16x8xf32>) 39// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_6]] to %[[VAL_3]] step %[[VAL_7]] { 40// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 41// CHECK: %[[VAL_14:.*]] = arith.muli %[[VAL_12]], %[[VAL_4]] : index 42// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_14]], %[[VAL_13]] : index 43// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 44// CHECK: %[[VAL_17:.*]] = arith.muli %[[VAL_15]], %[[VAL_5]] : index 45// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_17]], %[[VAL_16]] : index 46// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xf32> 47// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 48// CHECK: %[[VAL_21:.*]] = arith.addf %[[VAL_19]], %[[VAL_20]] : f32 49// CHECK: memref.store %[[VAL_21]], %[[VAL_11]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 50// CHECK: } 51// CHECK: } 52// CHECK: } 53// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_11]] : memref<32x16x8xf32> 54// CHECK: return %[[VAL_22]] : tensor<32x16x8xf32> 55// CHECK: } 56func.func @add_ddd(%arga: tensor<32x16x8xf32, #Tddd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 57 %0 = linalg.generic #trait3 58 ins(%arga, %argb: tensor<32x16x8xf32, #Tddd>, tensor<32x16x8xf32>) 59 outs(%argx: tensor<32x16x8xf32>) { 60 ^bb(%a: f32, %b: f32, %x: f32): 61 %0 = arith.addf %a, %b : f32 62 linalg.yield %0 : f32 63 } -> tensor<32x16x8xf32> 64 return %0 : tensor<32x16x8xf32> 65} 66 67// CHECK-LABEL: func @mul_ddd( 68// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 69// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 70// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 71// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 72// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 73// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 74// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 75// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 76// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 77// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 78// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 79// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 80// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_11]] : memref<32x16x8xf32>) 81// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_6]] to %[[VAL_3]] step %[[VAL_7]] { 82// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 83// CHECK: %[[VAL_14:.*]] = arith.muli %[[VAL_12]], %[[VAL_4]] : index 84// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_14]], %[[VAL_13]] : index 85// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 86// CHECK: %[[VAL_17:.*]] = arith.muli %[[VAL_15]], %[[VAL_5]] : index 87// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_17]], %[[VAL_16]] : index 88// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xf32> 89// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 90// CHECK: %[[VAL_21:.*]] = arith.mulf %[[VAL_19]], %[[VAL_20]] : f32 91// CHECK: memref.store %[[VAL_21]], %[[VAL_11]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 92// CHECK: } 93// CHECK: } 94// CHECK: } 95// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_11]] : memref<32x16x8xf32> 96// CHECK: return %[[VAL_22]] : tensor<32x16x8xf32> 97// CHECK: } 98func.func @mul_ddd(%arga: tensor<32x16x8xf32, #Tddd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 99 %0 = linalg.generic #trait3 100 ins(%arga, %argb: tensor<32x16x8xf32, #Tddd>, tensor<32x16x8xf32>) 101 outs(%argx: tensor<32x16x8xf32>) { 102 ^bb(%a: f32, %b: f32, %x: f32): 103 %0 = arith.mulf %a, %b : f32 104 linalg.yield %0 : f32 105 } -> tensor<32x16x8xf32> 106 return %0 : tensor<32x16x8xf32> 107} 108 109// CHECK-LABEL: func @add_dds( 110// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 111// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 112// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 113// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 114// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 115// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 116// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 117// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 118// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 119// CHECK-DAG: %[[VAL_8:.*]] = arith.constant true 120// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 121// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 122// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 123// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 124// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 125// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 126// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_15]] : memref<32x16x8xf32>) 127// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_9]] { 128// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_9]] { 129// CHECK: %[[VAL_18:.*]] = arith.muli %[[VAL_16]], %[[VAL_5]] : index 130// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_18]], %[[VAL_17]] : index 131// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_19]]] : memref<?xindex> 132// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_19]], %[[VAL_9]] : index 133// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex> 134// CHECK: %[[VAL_23:.*]]:2 = scf.while (%[[VAL_24:.*]] = %[[VAL_20]], %[[VAL_25:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 135// CHECK: %[[VAL_26:.*]] = arith.cmpi ult, %[[VAL_24]], %[[VAL_22]] : index 136// CHECK: scf.condition(%[[VAL_26]]) %[[VAL_24]], %[[VAL_25]] : index, index 137// CHECK: } do { 138// CHECK: ^bb0(%[[VAL_27:.*]]: index, %[[VAL_28:.*]]: index): 139// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_27]]] : memref<?xindex> 140// CHECK: %[[VAL_30:.*]] = arith.cmpi eq, %[[VAL_29]], %[[VAL_28]] : index 141// CHECK: scf.if %[[VAL_30]] { 142// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_27]]] : memref<?xf32> 143// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 144// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 145// CHECK: memref.store %[[VAL_33]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 146// CHECK: } else { 147// CHECK: scf.if %[[VAL_8]] { 148// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 149// CHECK: memref.store %[[VAL_34]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 150// CHECK: } else { 151// CHECK: } 152// CHECK: } 153// CHECK: %[[VAL_35:.*]] = arith.cmpi eq, %[[VAL_29]], %[[VAL_28]] : index 154// CHECK: %[[VAL_36:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 155// CHECK: %[[VAL_37:.*]] = arith.select %[[VAL_35]], %[[VAL_36]], %[[VAL_27]] : index 156// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_28]], %[[VAL_9]] : index 157// CHECK: scf.yield %[[VAL_37]], %[[VAL_38]] : index, index 158// CHECK: } 159// CHECK: scf.for %[[VAL_39:.*]] = %[[VAL_40:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 160// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_39]]] : memref<32x16x8xf32> 161// CHECK: memref.store %[[VAL_41]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_39]]] : memref<32x16x8xf32> 162// CHECK: } 163// CHECK: } 164// CHECK: } 165// CHECK: %[[VAL_42:.*]] = bufferization.to_tensor %[[VAL_15]] : memref<32x16x8xf32> 166// CHECK: return %[[VAL_42]] : tensor<32x16x8xf32> 167// CHECK: } 168func.func @add_dds(%arga: tensor<32x16x8xf32, #Tdds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 169 %0 = linalg.generic #trait3 170 ins(%arga, %argb: tensor<32x16x8xf32, #Tdds>, tensor<32x16x8xf32>) 171 outs(%argx: tensor<32x16x8xf32>) { 172 ^bb(%a: f32, %b: f32, %x: f32): 173 %0 = arith.addf %a, %b : f32 174 linalg.yield %0 : f32 175 } -> tensor<32x16x8xf32> 176 return %0 : tensor<32x16x8xf32> 177} 178 179// CHECK-LABEL: func @mul_dds( 180// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 181// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 182// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 183// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 184// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 185// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 186// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 187// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 188// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 189// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 190// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 191// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 192// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 193// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 194// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<32x16x8xf32>) 195// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 196// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 197// CHECK: %[[VAL_16:.*]] = arith.muli %[[VAL_14]], %[[VAL_5]] : index 198// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_16]], %[[VAL_15]] : index 199// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 200// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_17]], %[[VAL_7]] : index 201// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 202// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_7]] { 203// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 204// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xf32> 205// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_14]], %[[VAL_15]], %[[VAL_22]]] : memref<32x16x8xf32> 206// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_23]], %[[VAL_24]] : f32 207// CHECK: memref.store %[[VAL_25]], %[[VAL_13]]{{\[}}%[[VAL_14]], %[[VAL_15]], %[[VAL_22]]] : memref<32x16x8xf32> 208// CHECK: } 209// CHECK: } 210// CHECK: } 211// CHECK: %[[VAL_26:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<32x16x8xf32> 212// CHECK: return %[[VAL_26]] : tensor<32x16x8xf32> 213// CHECK: } 214func.func @mul_dds(%arga: tensor<32x16x8xf32, #Tdds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 215 %0 = linalg.generic #trait3 216 ins(%arga, %argb: tensor<32x16x8xf32, #Tdds>, tensor<32x16x8xf32>) 217 outs(%argx: tensor<32x16x8xf32>) { 218 ^bb(%a: f32, %b: f32, %x: f32): 219 %0 = arith.mulf %a, %b : f32 220 linalg.yield %0 : f32 221 } -> tensor<32x16x8xf32> 222 return %0 : tensor<32x16x8xf32> 223} 224 225// CHECK-LABEL: func @add_dsd( 226// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 227// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 228// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 229// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 230// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 231// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 232// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 233// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 234// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 235// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 236// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 237// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 238// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 239// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 240// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 241// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 242// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_7]] to %[[VAL_3]] step %[[VAL_8]] { 243// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_15]]] : memref<?xindex> 244// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_8]] : index 245// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_17]]] : memref<?xindex> 246// CHECK: %[[VAL_19:.*]]:2 = scf.while (%[[VAL_20:.*]] = %[[VAL_16]], %[[VAL_21:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 247// CHECK: %[[VAL_22:.*]] = arith.cmpi ult, %[[VAL_20]], %[[VAL_18]] : index 248// CHECK: scf.condition(%[[VAL_22]]) %[[VAL_20]], %[[VAL_21]] : index, index 249// CHECK: } do { 250// CHECK: ^bb0(%[[VAL_23:.*]]: index, %[[VAL_24:.*]]: index): 251// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 252// CHECK: %[[VAL_26:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 253// CHECK: scf.if %[[VAL_26]] { 254// CHECK: scf.for %[[VAL_27:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 255// CHECK: %[[VAL_28:.*]] = arith.muli %[[VAL_23]], %[[VAL_5]] : index 256// CHECK: %[[VAL_29:.*]] = arith.addi %[[VAL_28]], %[[VAL_27]] : index 257// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_29]]] : memref<?xf32> 258// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_27]]] : memref<32x16x8xf32> 259// CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[VAL_31]] : f32 260// CHECK: memref.store %[[VAL_32]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_27]]] : memref<32x16x8xf32> 261// CHECK: } 262// CHECK: } else { 263// CHECK: scf.if %[[VAL_6]] { 264// CHECK: scf.for %[[VAL_33:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 265// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_33]]] : memref<32x16x8xf32> 266// CHECK: memref.store %[[VAL_34]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_33]]] : memref<32x16x8xf32> 267// CHECK: } 268// CHECK: } else { 269// CHECK: } 270// CHECK: } 271// CHECK: %[[VAL_35:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 272// CHECK: %[[VAL_36:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 273// CHECK: %[[VAL_37:.*]] = arith.select %[[VAL_35]], %[[VAL_36]], %[[VAL_23]] : index 274// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_24]], %[[VAL_8]] : index 275// CHECK: scf.yield %[[VAL_37]], %[[VAL_38]] : index, index 276// CHECK: } 277// CHECK: scf.for %[[VAL_39:.*]] = %[[VAL_40:.*]]#1 to %[[VAL_4]] step %[[VAL_8]] { 278// CHECK: scf.for %[[VAL_41:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 279// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_39]], %[[VAL_41]]] : memref<32x16x8xf32> 280// CHECK: memref.store %[[VAL_42]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_39]], %[[VAL_41]]] : memref<32x16x8xf32> 281// CHECK: } 282// CHECK: } 283// CHECK: } 284// CHECK: %[[VAL_43:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 285// CHECK: return %[[VAL_43]] : tensor<32x16x8xf32> 286// CHECK: } 287func.func @add_dsd(%arga: tensor<32x16x8xf32, #Tdsd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 288 %0 = linalg.generic #trait3 289 ins(%arga, %argb: tensor<32x16x8xf32, #Tdsd>, tensor<32x16x8xf32>) 290 outs(%argx: tensor<32x16x8xf32>) { 291 ^bb(%a: f32, %b: f32, %x: f32): 292 %0 = arith.addf %a, %b : f32 293 linalg.yield %0 : f32 294 } -> tensor<32x16x8xf32> 295 return %0 : tensor<32x16x8xf32> 296} 297 298// CHECK-LABEL: func @mul_dsd( 299// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 300// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 301// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 302// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 303// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 304// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 8 : index 305// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 306// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 307// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_6]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 308// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_6]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 309// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 310// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 311// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 312// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_12]] : memref<32x16x8xf32>) 313// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { 314// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_13]]] : memref<?xindex> 315// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_13]], %[[VAL_6]] : index 316// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex> 317// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_14]] to %[[VAL_16]] step %[[VAL_6]] { 318// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 319// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 320// CHECK: %[[VAL_20:.*]] = arith.muli %[[VAL_17]], %[[VAL_4]] : index 321// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_20]], %[[VAL_19]] : index 322// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xf32> 323// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_13]], %[[VAL_18]], %[[VAL_19]]] : memref<32x16x8xf32> 324// CHECK: %[[VAL_24:.*]] = arith.mulf %[[VAL_22]], %[[VAL_23]] : f32 325// CHECK: memref.store %[[VAL_24]], %[[VAL_12]]{{\[}}%[[VAL_13]], %[[VAL_18]], %[[VAL_19]]] : memref<32x16x8xf32> 326// CHECK: } 327// CHECK: } 328// CHECK: } 329// CHECK: %[[VAL_25:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<32x16x8xf32> 330// CHECK: return %[[VAL_25]] : tensor<32x16x8xf32> 331// CHECK: } 332func.func @mul_dsd(%arga: tensor<32x16x8xf32, #Tdsd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 333 %0 = linalg.generic #trait3 334 ins(%arga, %argb: tensor<32x16x8xf32, #Tdsd>, tensor<32x16x8xf32>) 335 outs(%argx: tensor<32x16x8xf32>) { 336 ^bb(%a: f32, %b: f32, %x: f32): 337 %0 = arith.mulf %a, %b : f32 338 linalg.yield %0 : f32 339 } -> tensor<32x16x8xf32> 340 return %0 : tensor<32x16x8xf32> 341} 342 343// CHECK-LABEL: func @add_dss( 344// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 345// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 346// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 347// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 348// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 349// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 350// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 351// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 352// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 353// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 354// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 355// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_9]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 356// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_9]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 357// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 358// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 359// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 360// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 361// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 362// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_17]] : memref<32x16x8xf32>) 363// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_8]] to %[[VAL_4]] step %[[VAL_9]] { 364// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_18]]] : memref<?xindex> 365// CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_18]], %[[VAL_9]] : index 366// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_20]]] : memref<?xindex> 367// CHECK: %[[VAL_22:.*]]:2 = scf.while (%[[VAL_23:.*]] = %[[VAL_19]], %[[VAL_24:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 368// CHECK: %[[VAL_25:.*]] = arith.cmpi ult, %[[VAL_23]], %[[VAL_21]] : index 369// CHECK: scf.condition(%[[VAL_25]]) %[[VAL_23]], %[[VAL_24]] : index, index 370// CHECK: } do { 371// CHECK: ^bb0(%[[VAL_26:.*]]: index, %[[VAL_27:.*]]: index): 372// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_26]]] : memref<?xindex> 373// CHECK: %[[VAL_29:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 374// CHECK: scf.if %[[VAL_29]] { 375// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_26]]] : memref<?xindex> 376// CHECK: %[[VAL_31:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 377// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_31]]] : memref<?xindex> 378// CHECK: %[[VAL_33:.*]]:2 = scf.while (%[[VAL_34:.*]] = %[[VAL_30]], %[[VAL_35:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 379// CHECK: %[[VAL_36:.*]] = arith.cmpi ult, %[[VAL_34]], %[[VAL_32]] : index 380// CHECK: scf.condition(%[[VAL_36]]) %[[VAL_34]], %[[VAL_35]] : index, index 381// CHECK: } do { 382// CHECK: ^bb0(%[[VAL_37:.*]]: index, %[[VAL_38:.*]]: index): 383// CHECK: %[[VAL_39:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_37]]] : memref<?xindex> 384// CHECK: %[[VAL_40:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 385// CHECK: scf.if %[[VAL_40]] { 386// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_37]]] : memref<?xf32> 387// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 388// CHECK: %[[VAL_43:.*]] = arith.addf %[[VAL_41]], %[[VAL_42]] : f32 389// CHECK: memref.store %[[VAL_43]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 390// CHECK: } else { 391// CHECK: scf.if %[[VAL_7]] { 392// CHECK: %[[VAL_44:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 393// CHECK: memref.store %[[VAL_44]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 394// CHECK: } else { 395// CHECK: } 396// CHECK: } 397// CHECK: %[[VAL_45:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 398// CHECK: %[[VAL_46:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 399// CHECK: %[[VAL_47:.*]] = arith.select %[[VAL_45]], %[[VAL_46]], %[[VAL_37]] : index 400// CHECK: %[[VAL_48:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 401// CHECK: scf.yield %[[VAL_47]], %[[VAL_48]] : index, index 402// CHECK: } 403// CHECK: scf.for %[[VAL_49:.*]] = %[[VAL_50:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 404// CHECK: %[[VAL_51:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_49]]] : memref<32x16x8xf32> 405// CHECK: memref.store %[[VAL_51]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_49]]] : memref<32x16x8xf32> 406// CHECK: } 407// CHECK: } else { 408// CHECK: scf.if %[[VAL_7]] { 409// CHECK: scf.for %[[VAL_52:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 410// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_52]]] : memref<32x16x8xf32> 411// CHECK: memref.store %[[VAL_53]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_52]]] : memref<32x16x8xf32> 412// CHECK: } 413// CHECK: } else { 414// CHECK: } 415// CHECK: } 416// CHECK: %[[VAL_54:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 417// CHECK: %[[VAL_55:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 418// CHECK: %[[VAL_56:.*]] = arith.select %[[VAL_54]], %[[VAL_55]], %[[VAL_26]] : index 419// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 420// CHECK: scf.yield %[[VAL_56]], %[[VAL_57]] : index, index 421// CHECK: } 422// CHECK: scf.for %[[VAL_58:.*]] = %[[VAL_59:.*]]#1 to %[[VAL_5]] step %[[VAL_9]] { 423// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 424// CHECK: %[[VAL_61:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_58]], %[[VAL_60]]] : memref<32x16x8xf32> 425// CHECK: memref.store %[[VAL_61]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_58]], %[[VAL_60]]] : memref<32x16x8xf32> 426// CHECK: } 427// CHECK: } 428// CHECK: } 429// CHECK: %[[VAL_62:.*]] = bufferization.to_tensor %[[VAL_17]] : memref<32x16x8xf32> 430// CHECK: return %[[VAL_62]] : tensor<32x16x8xf32> 431// CHECK: } 432func.func @add_dss(%arga: tensor<32x16x8xf32, #Tdss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 433 %0 = linalg.generic #trait3 434 ins(%arga, %argb: tensor<32x16x8xf32, #Tdss>, tensor<32x16x8xf32>) 435 outs(%argx: tensor<32x16x8xf32>) { 436 ^bb(%a: f32, %b: f32, %x: f32): 437 %0 = arith.addf %a, %b : f32 438 linalg.yield %0 : f32 439 } -> tensor<32x16x8xf32> 440 return %0 : tensor<32x16x8xf32> 441} 442 443// CHECK-LABEL: func @mul_dss( 444// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 445// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 446// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 447// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 448// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 449// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 450// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 451// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 452// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_6]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 453// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_6]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 454// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 455// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 456// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 457// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 458// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 459// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 460// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 461// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex> 462// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_6]] : index 463// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_17]]] : memref<?xindex> 464// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_16]] to %[[VAL_18]] step %[[VAL_6]] { 465// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 466// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_19]]] : memref<?xindex> 467// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_19]], %[[VAL_6]] : index 468// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_22]]] : memref<?xindex> 469// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_21]] to %[[VAL_23]] step %[[VAL_6]] { 470// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_24]]] : memref<?xindex> 471// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_24]]] : memref<?xf32> 472// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_20]], %[[VAL_25]]] : memref<32x16x8xf32> 473// CHECK: %[[VAL_28:.*]] = arith.mulf %[[VAL_26]], %[[VAL_27]] : f32 474// CHECK: memref.store %[[VAL_28]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_20]], %[[VAL_25]]] : memref<32x16x8xf32> 475// CHECK: } 476// CHECK: } 477// CHECK: } 478// CHECK: %[[VAL_29:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 479// CHECK: return %[[VAL_29]] : tensor<32x16x8xf32> 480// CHECK: } 481func.func @mul_dss(%arga: tensor<32x16x8xf32, #Tdss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 482 %0 = linalg.generic #trait3 483 ins(%arga, %argb: tensor<32x16x8xf32, #Tdss>, tensor<32x16x8xf32>) 484 outs(%argx: tensor<32x16x8xf32>) { 485 ^bb(%a: f32, %b: f32, %x: f32): 486 %0 = arith.mulf %a, %b : f32 487 linalg.yield %0 : f32 488 } -> tensor<32x16x8xf32> 489 return %0 : tensor<32x16x8xf32> 490} 491 492// CHECK-LABEL: func @add_sdd( 493// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 494// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 495// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 496// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 497// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 498// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 499// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 500// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 501// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 502// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 503// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_7]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 504// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_7]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 505// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 506// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 507// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 508// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 509// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_7]]] : memref<?xindex> 510// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_8]]] : memref<?xindex> 511// CHECK: %[[VAL_17:.*]]:2 = scf.while (%[[VAL_18:.*]] = %[[VAL_15]], %[[VAL_19:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 512// CHECK: %[[VAL_20:.*]] = arith.cmpi ult, %[[VAL_18]], %[[VAL_16]] : index 513// CHECK: scf.condition(%[[VAL_20]]) %[[VAL_18]], %[[VAL_19]] : index, index 514// CHECK: } do { 515// CHECK: ^bb0(%[[VAL_21:.*]]: index, %[[VAL_22:.*]]: index): 516// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex> 517// CHECK: %[[VAL_24:.*]] = arith.cmpi eq, %[[VAL_23]], %[[VAL_22]] : index 518// CHECK: scf.if %[[VAL_24]] { 519// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 520// CHECK: %[[VAL_26:.*]] = arith.muli %[[VAL_21]], %[[VAL_4]] : index 521// CHECK: %[[VAL_27:.*]] = arith.addi %[[VAL_26]], %[[VAL_25]] : index 522// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 523// CHECK: %[[VAL_29:.*]] = arith.muli %[[VAL_27]], %[[VAL_5]] : index 524// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_29]], %[[VAL_28]] : index 525// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_30]]] : memref<?xf32> 526// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]], %[[VAL_25]], %[[VAL_28]]] : memref<32x16x8xf32> 527// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 528// CHECK: memref.store %[[VAL_33]], %[[VAL_14]]{{\[}}%[[VAL_22]], %[[VAL_25]], %[[VAL_28]]] : memref<32x16x8xf32> 529// CHECK: } 530// CHECK: } 531// CHECK: } else { 532// CHECK: scf.if %[[VAL_6]] { 533// CHECK: scf.for %[[VAL_34:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 534// CHECK: scf.for %[[VAL_35:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 535// CHECK: %[[VAL_36:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]], %[[VAL_34]], %[[VAL_35]]] : memref<32x16x8xf32> 536// CHECK: memref.store %[[VAL_36]], %[[VAL_14]]{{\[}}%[[VAL_22]], %[[VAL_34]], %[[VAL_35]]] : memref<32x16x8xf32> 537// CHECK: } 538// CHECK: } 539// CHECK: } else { 540// CHECK: } 541// CHECK: } 542// CHECK: %[[VAL_37:.*]] = arith.cmpi eq, %[[VAL_23]], %[[VAL_22]] : index 543// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_21]], %[[VAL_8]] : index 544// CHECK: %[[VAL_39:.*]] = arith.select %[[VAL_37]], %[[VAL_38]], %[[VAL_21]] : index 545// CHECK: %[[VAL_40:.*]] = arith.addi %[[VAL_22]], %[[VAL_8]] : index 546// CHECK: scf.yield %[[VAL_39]], %[[VAL_40]] : index, index 547// CHECK: } 548// CHECK: scf.for %[[VAL_41:.*]] = %[[VAL_42:.*]]#1 to %[[VAL_3]] step %[[VAL_8]] { 549// CHECK: scf.for %[[VAL_43:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 550// CHECK: scf.for %[[VAL_44:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 551// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_41]], %[[VAL_43]], %[[VAL_44]]] : memref<32x16x8xf32> 552// CHECK: memref.store %[[VAL_45]], %[[VAL_14]]{{\[}}%[[VAL_41]], %[[VAL_43]], %[[VAL_44]]] : memref<32x16x8xf32> 553// CHECK: } 554// CHECK: } 555// CHECK: } 556// CHECK: %[[VAL_46:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 557// CHECK: return %[[VAL_46]] : tensor<32x16x8xf32> 558// CHECK: } 559func.func @add_sdd(%arga: tensor<32x16x8xf32, #Tsdd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 560 %0 = linalg.generic #trait3 561 ins(%arga, %argb: tensor<32x16x8xf32, #Tsdd>, tensor<32x16x8xf32>) 562 outs(%argx: tensor<32x16x8xf32>) { 563 ^bb(%a: f32, %b: f32, %x: f32): 564 %0 = arith.addf %a, %b : f32 565 linalg.yield %0 : f32 566 } -> tensor<32x16x8xf32> 567 return %0 : tensor<32x16x8xf32> 568} 569 570// CHECK-LABEL: func @mul_sdd( 571// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 572// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 573// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 574// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 575// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 16 : index 576// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 8 : index 577// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 578// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 579// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 580// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 581// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 582// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 583// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 584// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_12]] : memref<32x16x8xf32>) 585// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_5]]] : memref<?xindex> 586// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_6]]] : memref<?xindex> 587// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_13]] to %[[VAL_14]] step %[[VAL_6]] { 588// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]]] : memref<?xindex> 589// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { 590// CHECK: %[[VAL_18:.*]] = arith.muli %[[VAL_15]], %[[VAL_3]] : index 591// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_18]], %[[VAL_17]] : index 592// CHECK: scf.for %[[VAL_20:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 593// CHECK: %[[VAL_21:.*]] = arith.muli %[[VAL_19]], %[[VAL_4]] : index 594// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_21]], %[[VAL_20]] : index 595// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_22]]] : memref<?xf32> 596// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_20]]] : memref<32x16x8xf32> 597// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_23]], %[[VAL_24]] : f32 598// CHECK: memref.store %[[VAL_25]], %[[VAL_12]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_20]]] : memref<32x16x8xf32> 599// CHECK: } 600// CHECK: } 601// CHECK: } 602// CHECK: %[[VAL_26:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<32x16x8xf32> 603// CHECK: return %[[VAL_26]] : tensor<32x16x8xf32> 604// CHECK: } 605func.func @mul_sdd(%arga: tensor<32x16x8xf32, #Tsdd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 606 %0 = linalg.generic #trait3 607 ins(%arga, %argb: tensor<32x16x8xf32, #Tsdd>, tensor<32x16x8xf32>) 608 outs(%argx: tensor<32x16x8xf32>) { 609 ^bb(%a: f32, %b: f32, %x: f32): 610 %0 = arith.mulf %a, %b : f32 611 linalg.yield %0 : f32 612 } -> tensor<32x16x8xf32> 613 return %0 : tensor<32x16x8xf32> 614} 615 616// CHECK-LABEL: func @add_sds( 617// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 618// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 619// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 620// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 621// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 622// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 623// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 624// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 625// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 626// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 627// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 628// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 629// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 630// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 631// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 632// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 633// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 634// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 635// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_17]] : memref<32x16x8xf32>) 636// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_8]]] : memref<?xindex> 637// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_9]]] : memref<?xindex> 638// CHECK: %[[VAL_20:.*]]:2 = scf.while (%[[VAL_21:.*]] = %[[VAL_18]], %[[VAL_22:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 639// CHECK: %[[VAL_23:.*]] = arith.cmpi ult, %[[VAL_21]], %[[VAL_19]] : index 640// CHECK: scf.condition(%[[VAL_23]]) %[[VAL_21]], %[[VAL_22]] : index, index 641// CHECK: } do { 642// CHECK: ^bb0(%[[VAL_24:.*]]: index, %[[VAL_25:.*]]: index): 643// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_24]]] : memref<?xindex> 644// CHECK: %[[VAL_27:.*]] = arith.cmpi eq, %[[VAL_26]], %[[VAL_25]] : index 645// CHECK: scf.if %[[VAL_27]] { 646// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 647// CHECK: %[[VAL_29:.*]] = arith.muli %[[VAL_24]], %[[VAL_5]] : index 648// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_29]], %[[VAL_28]] : index 649// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_30]]] : memref<?xindex> 650// CHECK: %[[VAL_32:.*]] = arith.addi %[[VAL_30]], %[[VAL_9]] : index 651// CHECK: %[[VAL_33:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_32]]] : memref<?xindex> 652// CHECK: %[[VAL_34:.*]]:2 = scf.while (%[[VAL_35:.*]] = %[[VAL_31]], %[[VAL_36:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 653// CHECK: %[[VAL_37:.*]] = arith.cmpi ult, %[[VAL_35]], %[[VAL_33]] : index 654// CHECK: scf.condition(%[[VAL_37]]) %[[VAL_35]], %[[VAL_36]] : index, index 655// CHECK: } do { 656// CHECK: ^bb0(%[[VAL_38:.*]]: index, %[[VAL_39:.*]]: index): 657// CHECK: %[[VAL_40:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_38]]] : memref<?xindex> 658// CHECK: %[[VAL_41:.*]] = arith.cmpi eq, %[[VAL_40]], %[[VAL_39]] : index 659// CHECK: scf.if %[[VAL_41]] { 660// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_38]]] : memref<?xf32> 661// CHECK: %[[VAL_43:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 662// CHECK: %[[VAL_44:.*]] = arith.addf %[[VAL_42]], %[[VAL_43]] : f32 663// CHECK: memref.store %[[VAL_44]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 664// CHECK: } else { 665// CHECK: scf.if %[[VAL_7]] { 666// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 667// CHECK: memref.store %[[VAL_45]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 668// CHECK: } else { 669// CHECK: } 670// CHECK: } 671// CHECK: %[[VAL_46:.*]] = arith.cmpi eq, %[[VAL_40]], %[[VAL_39]] : index 672// CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 673// CHECK: %[[VAL_48:.*]] = arith.select %[[VAL_46]], %[[VAL_47]], %[[VAL_38]] : index 674// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_39]], %[[VAL_9]] : index 675// CHECK: scf.yield %[[VAL_48]], %[[VAL_49]] : index, index 676// CHECK: } 677// CHECK: scf.for %[[VAL_50:.*]] = %[[VAL_51:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 678// CHECK: %[[VAL_52:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_50]]] : memref<32x16x8xf32> 679// CHECK: memref.store %[[VAL_52]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_50]]] : memref<32x16x8xf32> 680// CHECK: } 681// CHECK: } 682// CHECK: } else { 683// CHECK: scf.if %[[VAL_7]] { 684// CHECK: scf.for %[[VAL_53:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 685// CHECK: scf.for %[[VAL_54:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 686// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_53]], %[[VAL_54]]] : memref<32x16x8xf32> 687// CHECK: memref.store %[[VAL_55]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_53]], %[[VAL_54]]] : memref<32x16x8xf32> 688// CHECK: } 689// CHECK: } 690// CHECK: } else { 691// CHECK: } 692// CHECK: } 693// CHECK: %[[VAL_56:.*]] = arith.cmpi eq, %[[VAL_26]], %[[VAL_25]] : index 694// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_24]], %[[VAL_9]] : index 695// CHECK: %[[VAL_58:.*]] = arith.select %[[VAL_56]], %[[VAL_57]], %[[VAL_24]] : index 696// CHECK: %[[VAL_59:.*]] = arith.addi %[[VAL_25]], %[[VAL_9]] : index 697// CHECK: scf.yield %[[VAL_58]], %[[VAL_59]] : index, index 698// CHECK: } 699// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_61:.*]]#1 to %[[VAL_4]] step %[[VAL_9]] { 700// CHECK: scf.for %[[VAL_62:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 701// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 702// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_60]], %[[VAL_62]], %[[VAL_63]]] : memref<32x16x8xf32> 703// CHECK: memref.store %[[VAL_64]], %[[VAL_17]]{{\[}}%[[VAL_60]], %[[VAL_62]], %[[VAL_63]]] : memref<32x16x8xf32> 704// CHECK: } 705// CHECK: } 706// CHECK: } 707// CHECK: %[[VAL_65:.*]] = bufferization.to_tensor %[[VAL_17]] : memref<32x16x8xf32> 708// CHECK: return %[[VAL_65]] : tensor<32x16x8xf32> 709// CHECK: } 710func.func @add_sds(%arga: tensor<32x16x8xf32, #Tsds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 711 %0 = linalg.generic #trait3 712 ins(%arga, %argb: tensor<32x16x8xf32, #Tsds>, tensor<32x16x8xf32>) 713 outs(%argx: tensor<32x16x8xf32>) { 714 ^bb(%a: f32, %b: f32, %x: f32): 715 %0 = arith.addf %a, %b : f32 716 linalg.yield %0 : f32 717 } -> tensor<32x16x8xf32> 718 return %0 : tensor<32x16x8xf32> 719} 720 721// CHECK-LABEL: func @mul_sds( 722// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 723// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 724// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 725// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 726// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 727// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 728// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 729// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 730// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 731// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 732// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 733// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 734// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 735// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 736// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 737// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 738// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_5]]] : memref<?xindex> 739// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_6]]] : memref<?xindex> 740// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_15]] to %[[VAL_16]] step %[[VAL_6]] { 741// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 742// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 743// CHECK: %[[VAL_20:.*]] = arith.muli %[[VAL_17]], %[[VAL_4]] : index 744// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_20]], %[[VAL_19]] : index 745// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 746// CHECK: %[[VAL_23:.*]] = arith.addi %[[VAL_21]], %[[VAL_6]] : index 747// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_23]]] : memref<?xindex> 748// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_22]] to %[[VAL_24]] step %[[VAL_6]] { 749// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_25]]] : memref<?xindex> 750// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_25]]] : memref<?xf32> 751// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_18]], %[[VAL_19]], %[[VAL_26]]] : memref<32x16x8xf32> 752// CHECK: %[[VAL_29:.*]] = arith.mulf %[[VAL_27]], %[[VAL_28]] : f32 753// CHECK: memref.store %[[VAL_29]], %[[VAL_14]]{{\[}}%[[VAL_18]], %[[VAL_19]], %[[VAL_26]]] : memref<32x16x8xf32> 754// CHECK: } 755// CHECK: } 756// CHECK: } 757// CHECK: %[[VAL_30:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 758// CHECK: return %[[VAL_30]] : tensor<32x16x8xf32> 759// CHECK: } 760func.func @mul_sds(%arga: tensor<32x16x8xf32, #Tsds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 761 %0 = linalg.generic #trait3 762 ins(%arga, %argb: tensor<32x16x8xf32, #Tsds>, tensor<32x16x8xf32>) 763 outs(%argx: tensor<32x16x8xf32>) { 764 ^bb(%a: f32, %b: f32, %x: f32): 765 %0 = arith.mulf %a, %b : f32 766 linalg.yield %0 : f32 767 } -> tensor<32x16x8xf32> 768 return %0 : tensor<32x16x8xf32> 769} 770 771// CHECK-LABEL: func @add_ssd( 772// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 773// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 774// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 775// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 776// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 777// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 778// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 779// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 780// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 781// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 782// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_7]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 783// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_7]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 784// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 785// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 786// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 787// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 788// CHECK-DAG: %[[VAL_16:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 789// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_16]] : memref<32x16x8xf32>) 790// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_7]]] : memref<?xindex> 791// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_8]]] : memref<?xindex> 792// CHECK: %[[VAL_19:.*]]:2 = scf.while (%[[VAL_20:.*]] = %[[VAL_17]], %[[VAL_21:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 793// CHECK: %[[VAL_22:.*]] = arith.cmpi ult, %[[VAL_20]], %[[VAL_18]] : index 794// CHECK: scf.condition(%[[VAL_22]]) %[[VAL_20]], %[[VAL_21]] : index, index 795// CHECK: } do { 796// CHECK: ^bb0(%[[VAL_23:.*]]: index, %[[VAL_24:.*]]: index): 797// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 798// CHECK: %[[VAL_26:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 799// CHECK: scf.if %[[VAL_26]] { 800// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_23]]] : memref<?xindex> 801// CHECK: %[[VAL_28:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 802// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_28]]] : memref<?xindex> 803// CHECK: %[[VAL_30:.*]]:2 = scf.while (%[[VAL_31:.*]] = %[[VAL_27]], %[[VAL_32:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 804// CHECK: %[[VAL_33:.*]] = arith.cmpi ult, %[[VAL_31]], %[[VAL_29]] : index 805// CHECK: scf.condition(%[[VAL_33]]) %[[VAL_31]], %[[VAL_32]] : index, index 806// CHECK: } do { 807// CHECK: ^bb0(%[[VAL_34:.*]]: index, %[[VAL_35:.*]]: index): 808// CHECK: %[[VAL_36:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_34]]] : memref<?xindex> 809// CHECK: %[[VAL_37:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_35]] : index 810// CHECK: scf.if %[[VAL_37]] { 811// CHECK: scf.for %[[VAL_38:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 812// CHECK: %[[VAL_39:.*]] = arith.muli %[[VAL_34]], %[[VAL_5]] : index 813// CHECK: %[[VAL_40:.*]] = arith.addi %[[VAL_39]], %[[VAL_38]] : index 814// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_40]]] : memref<?xf32> 815// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_38]]] : memref<32x16x8xf32> 816// CHECK: %[[VAL_43:.*]] = arith.addf %[[VAL_41]], %[[VAL_42]] : f32 817// CHECK: memref.store %[[VAL_43]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_38]]] : memref<32x16x8xf32> 818// CHECK: } 819// CHECK: } else { 820// CHECK: scf.if %[[VAL_6]] { 821// CHECK: scf.for %[[VAL_44:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 822// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_44]]] : memref<32x16x8xf32> 823// CHECK: memref.store %[[VAL_45]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_44]]] : memref<32x16x8xf32> 824// CHECK: } 825// CHECK: } else { 826// CHECK: } 827// CHECK: } 828// CHECK: %[[VAL_46:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_35]] : index 829// CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_34]], %[[VAL_8]] : index 830// CHECK: %[[VAL_48:.*]] = arith.select %[[VAL_46]], %[[VAL_47]], %[[VAL_34]] : index 831// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_35]], %[[VAL_8]] : index 832// CHECK: scf.yield %[[VAL_48]], %[[VAL_49]] : index, index 833// CHECK: } 834// CHECK: scf.for %[[VAL_50:.*]] = %[[VAL_51:.*]]#1 to %[[VAL_4]] step %[[VAL_8]] { 835// CHECK: scf.for %[[VAL_52:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 836// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_50]], %[[VAL_52]]] : memref<32x16x8xf32> 837// CHECK: memref.store %[[VAL_53]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_50]], %[[VAL_52]]] : memref<32x16x8xf32> 838// CHECK: } 839// CHECK: } 840// CHECK: } else { 841// CHECK: scf.if %[[VAL_6]] { 842// CHECK: scf.for %[[VAL_54:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 843// CHECK: scf.for %[[VAL_55:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 844// CHECK: %[[VAL_56:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_54]], %[[VAL_55]]] : memref<32x16x8xf32> 845// CHECK: memref.store %[[VAL_56]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_54]], %[[VAL_55]]] : memref<32x16x8xf32> 846// CHECK: } 847// CHECK: } 848// CHECK: } else { 849// CHECK: } 850// CHECK: } 851// CHECK: %[[VAL_57:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 852// CHECK: %[[VAL_58:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 853// CHECK: %[[VAL_59:.*]] = arith.select %[[VAL_57]], %[[VAL_58]], %[[VAL_23]] : index 854// CHECK: %[[VAL_60:.*]] = arith.addi %[[VAL_24]], %[[VAL_8]] : index 855// CHECK: scf.yield %[[VAL_59]], %[[VAL_60]] : index, index 856// CHECK: } 857// CHECK: scf.for %[[VAL_61:.*]] = %[[VAL_62:.*]]#1 to %[[VAL_3]] step %[[VAL_8]] { 858// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 859// CHECK: scf.for %[[VAL_64:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 860// CHECK: %[[VAL_65:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_61]], %[[VAL_63]], %[[VAL_64]]] : memref<32x16x8xf32> 861// CHECK: memref.store %[[VAL_65]], %[[VAL_16]]{{\[}}%[[VAL_61]], %[[VAL_63]], %[[VAL_64]]] : memref<32x16x8xf32> 862// CHECK: } 863// CHECK: } 864// CHECK: } 865// CHECK: %[[VAL_66:.*]] = bufferization.to_tensor %[[VAL_16]] : memref<32x16x8xf32> 866// CHECK: return %[[VAL_66]] : tensor<32x16x8xf32> 867// CHECK: } 868func.func @add_ssd(%arga: tensor<32x16x8xf32, #Tssd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 869 %0 = linalg.generic #trait3 870 ins(%arga, %argb: tensor<32x16x8xf32, #Tssd>, tensor<32x16x8xf32>) 871 outs(%argx: tensor<32x16x8xf32>) { 872 ^bb(%a: f32, %b: f32, %x: f32): 873 %0 = arith.addf %a, %b : f32 874 linalg.yield %0 : f32 875 } -> tensor<32x16x8xf32> 876 return %0 : tensor<32x16x8xf32> 877} 878 879// CHECK-LABEL: func @mul_ssd( 880// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 881// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 882// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 883// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 884// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 8 : index 885// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index 886// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index 887// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 888// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 889// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 890// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 891// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 892// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 893// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 894// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<32x16x8xf32>) 895// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex> 896// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_5]]] : memref<?xindex> 897// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_5]] { 898// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xindex> 899// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex> 900// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_16]], %[[VAL_5]] : index 901// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 902// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_5]] { 903// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 904// CHECK: scf.for %[[VAL_23:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] { 905// CHECK: %[[VAL_24:.*]] = arith.muli %[[VAL_21]], %[[VAL_3]] : index 906// CHECK: %[[VAL_25:.*]] = arith.addi %[[VAL_24]], %[[VAL_23]] : index 907// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_25]]] : memref<?xf32> 908// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_17]], %[[VAL_22]], %[[VAL_23]]] : memref<32x16x8xf32> 909// CHECK: %[[VAL_28:.*]] = arith.mulf %[[VAL_26]], %[[VAL_27]] : f32 910// CHECK: memref.store %[[VAL_28]], %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_22]], %[[VAL_23]]] : memref<32x16x8xf32> 911// CHECK: } 912// CHECK: } 913// CHECK: } 914// CHECK: %[[VAL_29:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<32x16x8xf32> 915// CHECK: return %[[VAL_29]] : tensor<32x16x8xf32> 916// CHECK: } 917func.func @mul_ssd(%arga: tensor<32x16x8xf32, #Tssd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 918 %0 = linalg.generic #trait3 919 ins(%arga, %argb: tensor<32x16x8xf32, #Tssd>, tensor<32x16x8xf32>) 920 outs(%argx: tensor<32x16x8xf32>) { 921 ^bb(%a: f32, %b: f32, %x: f32): 922 %0 = arith.mulf %a, %b : f32 923 linalg.yield %0 : f32 924 } -> tensor<32x16x8xf32> 925 return %0 : tensor<32x16x8xf32> 926} 927 928// CHECK-LABEL: func @add_sss( 929// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 930// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 931// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 932// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 933// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 934// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 935// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 936// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 937// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 938// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 939// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 940// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 941// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_8]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 942// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_9]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 943// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_9]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 944// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 945// CHECK-DAG: %[[VAL_15:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 946// CHECK-DAG: %[[VAL_16:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 947// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 948// CHECK-DAG: %[[VAL_19:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 949// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_19]] : memref<32x16x8xf32>) 950// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_8]]] : memref<?xindex> 951// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_9]]] : memref<?xindex> 952// CHECK: %[[VAL_22:.*]]:2 = scf.while (%[[VAL_23:.*]] = %[[VAL_20]], %[[VAL_24:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 953// CHECK: %[[VAL_25:.*]] = arith.cmpi ult, %[[VAL_23]], %[[VAL_21]] : index 954// CHECK: scf.condition(%[[VAL_25]]) %[[VAL_23]], %[[VAL_24]] : index, index 955// CHECK: } do { 956// CHECK: ^bb0(%[[VAL_26:.*]]: index, %[[VAL_27:.*]]: index): 957// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_26]]] : memref<?xindex> 958// CHECK: %[[VAL_29:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 959// CHECK: scf.if %[[VAL_29]] { 960// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_26]]] : memref<?xindex> 961// CHECK: %[[VAL_31:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 962// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_31]]] : memref<?xindex> 963// CHECK: %[[VAL_33:.*]]:2 = scf.while (%[[VAL_34:.*]] = %[[VAL_30]], %[[VAL_35:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 964// CHECK: %[[VAL_36:.*]] = arith.cmpi ult, %[[VAL_34]], %[[VAL_32]] : index 965// CHECK: scf.condition(%[[VAL_36]]) %[[VAL_34]], %[[VAL_35]] : index, index 966// CHECK: } do { 967// CHECK: ^bb0(%[[VAL_37:.*]]: index, %[[VAL_38:.*]]: index): 968// CHECK: %[[VAL_39:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_37]]] : memref<?xindex> 969// CHECK: %[[VAL_40:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 970// CHECK: scf.if %[[VAL_40]] { 971// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_37]]] : memref<?xindex> 972// CHECK: %[[VAL_42:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 973// CHECK: %[[VAL_43:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_42]]] : memref<?xindex> 974// CHECK: %[[VAL_44:.*]]:2 = scf.while (%[[VAL_45:.*]] = %[[VAL_41]], %[[VAL_46:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 975// CHECK: %[[VAL_47:.*]] = arith.cmpi ult, %[[VAL_45]], %[[VAL_43]] : index 976// CHECK: scf.condition(%[[VAL_47]]) %[[VAL_45]], %[[VAL_46]] : index, index 977// CHECK: } do { 978// CHECK: ^bb0(%[[VAL_48:.*]]: index, %[[VAL_49:.*]]: index): 979// CHECK: %[[VAL_50:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_48]]] : memref<?xindex> 980// CHECK: %[[VAL_51:.*]] = arith.cmpi eq, %[[VAL_50]], %[[VAL_49]] : index 981// CHECK: scf.if %[[VAL_51]] { 982// CHECK: %[[VAL_52:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_48]]] : memref<?xf32> 983// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 984// CHECK: %[[VAL_54:.*]] = arith.addf %[[VAL_52]], %[[VAL_53]] : f32 985// CHECK: memref.store %[[VAL_54]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 986// CHECK: } else { 987// CHECK: scf.if %[[VAL_7]] { 988// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 989// CHECK: memref.store %[[VAL_55]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 990// CHECK: } else { 991// CHECK: } 992// CHECK: } 993// CHECK: %[[VAL_56:.*]] = arith.cmpi eq, %[[VAL_50]], %[[VAL_49]] : index 994// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_48]], %[[VAL_9]] : index 995// CHECK: %[[VAL_58:.*]] = arith.select %[[VAL_56]], %[[VAL_57]], %[[VAL_48]] : index 996// CHECK: %[[VAL_59:.*]] = arith.addi %[[VAL_49]], %[[VAL_9]] : index 997// CHECK: scf.yield %[[VAL_58]], %[[VAL_59]] : index, index 998// CHECK: } 999// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_61:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 1000// CHECK: %[[VAL_62:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_60]]] : memref<32x16x8xf32> 1001// CHECK: memref.store %[[VAL_62]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_60]]] : memref<32x16x8xf32> 1002// CHECK: } 1003// CHECK: } else { 1004// CHECK: scf.if %[[VAL_7]] { 1005// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1006// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_63]]] : memref<32x16x8xf32> 1007// CHECK: memref.store %[[VAL_64]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_63]]] : memref<32x16x8xf32> 1008// CHECK: } 1009// CHECK: } else { 1010// CHECK: } 1011// CHECK: } 1012// CHECK: %[[VAL_65:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 1013// CHECK: %[[VAL_66:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 1014// CHECK: %[[VAL_67:.*]] = arith.select %[[VAL_65]], %[[VAL_66]], %[[VAL_37]] : index 1015// CHECK: %[[VAL_68:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 1016// CHECK: scf.yield %[[VAL_67]], %[[VAL_68]] : index, index 1017// CHECK: } 1018// CHECK: scf.for %[[VAL_69:.*]] = %[[VAL_70:.*]]#1 to %[[VAL_5]] step %[[VAL_9]] { 1019// CHECK: scf.for %[[VAL_71:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1020// CHECK: %[[VAL_72:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_69]], %[[VAL_71]]] : memref<32x16x8xf32> 1021// CHECK: memref.store %[[VAL_72]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_69]], %[[VAL_71]]] : memref<32x16x8xf32> 1022// CHECK: } 1023// CHECK: } 1024// CHECK: } else { 1025// CHECK: scf.if %[[VAL_7]] { 1026// CHECK: scf.for %[[VAL_73:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 1027// CHECK: scf.for %[[VAL_74:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1028// CHECK: %[[VAL_75:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_73]], %[[VAL_74]]] : memref<32x16x8xf32> 1029// CHECK: memref.store %[[VAL_75]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_73]], %[[VAL_74]]] : memref<32x16x8xf32> 1030// CHECK: } 1031// CHECK: } 1032// CHECK: } else { 1033// CHECK: } 1034// CHECK: } 1035// CHECK: %[[VAL_76:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 1036// CHECK: %[[VAL_77:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 1037// CHECK: %[[VAL_78:.*]] = arith.select %[[VAL_76]], %[[VAL_77]], %[[VAL_26]] : index 1038// CHECK: %[[VAL_79:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 1039// CHECK: scf.yield %[[VAL_78]], %[[VAL_79]] : index, index 1040// CHECK: } 1041// CHECK: scf.for %[[VAL_80:.*]] = %[[VAL_81:.*]]#1 to %[[VAL_4]] step %[[VAL_9]] { 1042// CHECK: scf.for %[[VAL_82:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 1043// CHECK: scf.for %[[VAL_83:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1044// CHECK: %[[VAL_84:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_80]], %[[VAL_82]], %[[VAL_83]]] : memref<32x16x8xf32> 1045// CHECK: memref.store %[[VAL_84]], %[[VAL_19]]{{\[}}%[[VAL_80]], %[[VAL_82]], %[[VAL_83]]] : memref<32x16x8xf32> 1046// CHECK: } 1047// CHECK: } 1048// CHECK: } 1049// CHECK: %[[VAL_85:.*]] = bufferization.to_tensor %[[VAL_19]] : memref<32x16x8xf32> 1050// CHECK: return %[[VAL_85]] : tensor<32x16x8xf32> 1051// CHECK: } 1052func.func @add_sss(%arga: tensor<32x16x8xf32, #Tsss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1053 %0 = linalg.generic #trait3 1054 ins(%arga, %argb: tensor<32x16x8xf32, #Tsss>, tensor<32x16x8xf32>) 1055 outs(%argx: tensor<32x16x8xf32>) { 1056 ^bb(%a: f32, %b: f32, %x: f32): 1057 %0 = arith.addf %a, %b : f32 1058 linalg.yield %0 : f32 1059 } -> tensor<32x16x8xf32> 1060 return %0 : tensor<32x16x8xf32> 1061} 1062 1063// CHECK-LABEL: func @mul_sss( 1064// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 1065// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 1066// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1067// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 1068// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 2 : index 1069// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index 1070// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index 1071// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1072// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1073// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1074// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_5]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1075// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1076// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1077// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 1078// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_1]] : memref<32x16x8xf32> 1079// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_2]] : memref<32x16x8xf32> 1080// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_15]] : memref<32x16x8xf32>) 1081// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex> 1082// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_5]]] : memref<?xindex> 1083// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_16]] to %[[VAL_17]] step %[[VAL_5]] { 1084// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1085// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1086// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_18]], %[[VAL_5]] : index 1087// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_21]]] : memref<?xindex> 1088// CHECK: scf.for %[[VAL_23:.*]] = %[[VAL_20]] to %[[VAL_22]] step %[[VAL_5]] { 1089// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_23]]] : memref<?xindex> 1090// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 1091// CHECK: %[[VAL_26:.*]] = arith.addi %[[VAL_23]], %[[VAL_5]] : index 1092// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_26]]] : memref<?xindex> 1093// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_25]] to %[[VAL_27]] step %[[VAL_5]] { 1094// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_28]]] : memref<?xindex> 1095// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_28]]] : memref<?xf32> 1096// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_19]], %[[VAL_24]], %[[VAL_29]]] : memref<32x16x8xf32> 1097// CHECK: %[[VAL_32:.*]] = arith.mulf %[[VAL_30]], %[[VAL_31]] : f32 1098// CHECK: memref.store %[[VAL_32]], %[[VAL_15]]{{\[}}%[[VAL_19]], %[[VAL_24]], %[[VAL_29]]] : memref<32x16x8xf32> 1099// CHECK: } 1100// CHECK: } 1101// CHECK: } 1102// CHECK: %[[VAL_33:.*]] = bufferization.to_tensor %[[VAL_15]] : memref<32x16x8xf32> 1103// CHECK: return %[[VAL_33]] : tensor<32x16x8xf32> 1104// CHECK: } 1105func.func @mul_sss(%arga: tensor<32x16x8xf32, #Tsss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1106 %0 = linalg.generic #trait3 1107 ins(%arga, %argb: tensor<32x16x8xf32, #Tsss>, tensor<32x16x8xf32>) 1108 outs(%argx: tensor<32x16x8xf32>) { 1109 ^bb(%a: f32, %b: f32, %x: f32): 1110 %0 = arith.mulf %a, %b : f32 1111 linalg.yield %0 : f32 1112 } -> tensor<32x16x8xf32> 1113 return %0 : tensor<32x16x8xf32> 1114} 1115 1116#trait_kernel_3d = { 1117 indexing_maps = [ 1118 affine_map<(i,j,k,l) -> (i,k,l)>, // B 1119 affine_map<(i,j,k,l) -> (k,j)>, // C 1120 affine_map<(i,j,k,l) -> (l,j)>, // D 1121 affine_map<(i,j,k,l) -> (i,j)> // A (out) 1122 ], 1123 iterator_types = ["parallel", "parallel", "reduction", "reduction"], 1124 doc = "A(i,j) += SUM_k,l B(i,k,l) * C(k,j) * D(l,j)" 1125} 1126 1127// CHECK-LABEL: func @kernel_3d( 1128// CHECK-SAME: %[[VAL_0:.*0]]: tensor<?x?xf32>, 1129// CHECK-SAME: %[[VAL_1:.*1]]: tensor<?x?x?xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 1130// CHECK-SAME: %[[VAL_2:.*2]]: tensor<?x?xf32>, 1131// CHECK-SAME: %[[VAL_3:.*3]]: tensor<?x?xf32>) -> tensor<?x?xf32> { 1132// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 1133// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 1134// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 1135// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<?x?x?xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1136// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<?x?x?xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xindex> 1137// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?x?x?xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense", "dense", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 1138// CHECK-DAG: %[[VAL_10:.*]] = tensor.dim %[[VAL_2]], %[[VAL_5]] : tensor<?x?xf32> 1139// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : memref<?x?xf32> 1140// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_3]] : memref<?x?xf32> 1141// CHECK-DAG: %[[VAL_13:.*]] = tensor.dim %[[VAL_0]], %[[VAL_5]] : tensor<?x?xf32> 1142// CHECK-DAG: %[[VAL_14:.*]] = tensor.dim %[[VAL_0]], %[[VAL_6]] : tensor<?x?xf32> 1143// CHECK-DAG: %[[VAL_16:.*]] = bufferization.to_memref %[[VAL_0]] : memref<?x?xf32> 1144// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_5]] to %[[VAL_13]] step %[[VAL_6]] { 1145// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_5]] to %[[VAL_10]] step %[[VAL_6]] { 1146// CHECK: %[[VAL_19:.*]] = arith.muli %[[VAL_10]], %[[VAL_17]] : index 1147// CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_19]], %[[VAL_18]] : index 1148// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_20]]] : memref<?xindex> 1149// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_20]], %[[VAL_6]] : index 1150// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_22]]] : memref<?xindex> 1151// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_21]] to %[[VAL_23]] step %[[VAL_6]] { 1152// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_24]]] : memref<?xindex> 1153// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_24]]] : memref<?xf32> 1154// CHECK: scf.for %[[VAL_27:.*]] = %[[VAL_5]] to %[[VAL_14]] step %[[VAL_6]] { 1155// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_18]], %[[VAL_27]]] : memref<?x?xf32> 1156// CHECK: %[[VAL_29:.*]] = arith.mulf %[[VAL_26]], %[[VAL_28]] : f32 1157// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_25]], %[[VAL_27]]] : memref<?x?xf32> 1158// CHECK: %[[VAL_31:.*]] = arith.mulf %[[VAL_29]], %[[VAL_30]] : f32 1159// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_17]], %[[VAL_27]]] : memref<?x?xf32> 1160// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 1161// CHECK: memref.store %[[VAL_33]], %[[VAL_16]]{{\[}}%[[VAL_17]], %[[VAL_27]]] : memref<?x?xf32> 1162// CHECK: } 1163// CHECK: } 1164// CHECK: } 1165// CHECK: } 1166// CHECK: %[[VAL_34:.*]] = bufferization.to_tensor %[[VAL_16]] : memref<?x?xf32> 1167// CHECK: return %[[VAL_34]] : tensor<?x?xf32> 1168// CHECK: } 1169func.func @kernel_3d(%arga: tensor<?x?xf32>, 1170 %argb: tensor<?x?x?xf32, #Tdds>, 1171 %argc: tensor<?x?xf32>, 1172 %argd: tensor<?x?xf32>) -> tensor<?x?xf32> { 1173 %0 = linalg.generic #trait_kernel_3d 1174 ins(%argb, %argc, %argd: tensor<?x?x?xf32, #Tdds>, tensor<?x?xf32>, tensor<?x?xf32>) 1175 outs(%arga: tensor<?x?xf32>) { 1176 ^bb(%b: f32, %c: f32, %d: f32, %a: f32): 1177 %0 = arith.mulf %b, %c : f32 1178 %1 = arith.mulf %0, %d : f32 1179 %2 = arith.addf %1, %a : f32 1180 linalg.yield %2 : f32 1181 } -> tensor<?x?xf32> 1182 return %0 : tensor<?x?xf32> 1183} 1184 1185#trait_sum_reduction = { 1186 indexing_maps = [ 1187 affine_map<(i,j,k) -> (i,j,k)>, // A 1188 affine_map<(i,j,k) -> ()> // x (scalar out) 1189 ], 1190 iterator_types = ["reduction", "reduction", "reduction"], 1191 doc = "x += SUM_ijk A(i,j,k)" 1192} 1193 1194// CHECK-LABEL: func @sum_reduction( 1195// CHECK-SAME: %[[VAL_0:.*]]: tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}}>>, 1196// CHECK-SAME: %[[VAL_1:.*]]: tensor<f32>) -> tensor<f32> { 1197// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 1198// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 1199// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 1200// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_2]] : tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}>> 1201// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}>> 1202// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}}>> 1203// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}}>> 1204// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : memref<f32> 1205// CHECK: %[[VAL_11:.*]] = memref.load %[[VAL_10]][] : memref<f32> 1206// CHECK: %[[VAL_12:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_2]]] : memref<?xindex> 1207// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_3]]] : memref<?xindex> 1208// CHECK: %[[VAL_14:.*]] = scf.for %[[VAL_15:.*]] = %[[VAL_12]] to %[[VAL_13]] step %[[VAL_3]] iter_args(%[[VAL_16:.*]] = %[[VAL_11]]) -> (f32) { 1209// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_15]]] : memref<?xindex> 1210// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_15]], %[[VAL_3]] : index 1211// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1212// CHECK: %[[VAL_20:.*]] = scf.for %[[VAL_21:.*]] = %[[VAL_17]] to %[[VAL_19]] step %[[VAL_3]] iter_args(%[[VAL_22:.*]] = %[[VAL_16]]) -> (f32) { 1213// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_21]]] : memref<?xindex> 1214// CHECK: %[[VAL_24:.*]] = arith.addi %[[VAL_21]], %[[VAL_3]] : index 1215// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_24]]] : memref<?xindex> 1216// CHECK: %[[VAL_26:.*]] = scf.for %[[VAL_27:.*]] = %[[VAL_23]] to %[[VAL_25]] step %[[VAL_3]] iter_args(%[[VAL_28:.*]] = %[[VAL_22]]) -> (f32) { 1217// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_27]]] : memref<?xf32> 1218// CHECK: %[[VAL_30:.*]] = arith.addf %[[VAL_28]], %[[VAL_29]] : f32 1219// CHECK: scf.yield %[[VAL_30]] : f32 1220// CHECK: } 1221// CHECK: scf.yield %[[VAL_26]] : f32 1222// CHECK: } 1223// CHECK: scf.yield %[[VAL_20]] : f32 1224// CHECK: } 1225// CHECK: memref.store %[[VAL_14]], %[[VAL_10]][] : memref<f32> 1226// CHECK: %[[VAL_34:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<f32> 1227// CHECK: return %[[VAL_34]] : tensor<f32> 1228// CHECK: } 1229func.func @sum_reduction(%arga: tensor<10x20x30xf32, #Tsss>, %argx: tensor<f32>) -> tensor<f32> { 1230 %0 = linalg.generic #trait_sum_reduction 1231 ins(%arga: tensor<10x20x30xf32, #Tsss>) 1232 outs(%argx: tensor<f32>) { 1233 ^bb(%a: f32, %x: f32): 1234 %0 = arith.addf %x, %a : f32 1235 linalg.yield %0 : f32 1236 } -> tensor<f32> 1237 return %0 : tensor<f32> 1238} 1239 1240#trait_sum_reduction_inv = { 1241 indexing_maps = [ 1242 affine_map<(i,j,k) -> (i,j,k)>, // A 1243 affine_map<(i,j,k) -> (i)>, // b 1244 affine_map<(i,j,k) -> ()> // x (scalar out) 1245 ], 1246 iterator_types = ["reduction", "reduction", "reduction"], 1247 doc = "x += SUM_i A(i,j,k) * b(i)" 1248} 1249 1250// CHECK-LABEL: func @sum_reduction_inv( 1251// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?x?xf32>, 1252// CHECK-SAME: %[[VAL_1:.*]]: tensor<?xf32, #sparse_tensor.encoding<{{{.*}}}>> 1253// CHECK-SAME: %[[VAL_2:.*]]: tensor<f32>) -> tensor<f32> { 1254// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 1255// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 1256// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 1257// CHECK-DAG: %[[VAL_6:.*]] = tensor.dim %[[VAL_0]], %[[VAL_3]] : tensor<?x?x?xf32> 1258// CHECK-DAG: %[[VAL_7:.*]] = tensor.dim %[[VAL_0]], %[[VAL_4]] : tensor<?x?x?xf32> 1259// CHECK-DAG: %[[VAL_8:.*]] = bufferization.to_memref %[[VAL_0]] : memref<?x?x?xf32> 1260// CHECK-DAG: %[[VAL_9:.*]] = tensor.dim %[[VAL_1]], %[[VAL_5]] : tensor<?xf32, #sparse_tensor.encoding<{{{.*}}}>> 1261// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?xf32, #sparse_tensor.encoding<{{{.*}}}>> 1262// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : memref<f32> 1263// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_12]][] : memref<f32> 1264// CHECK: %[[VAL_14:.*]] = scf.for %[[VAL_15:.*]] = %[[VAL_5]] to %[[VAL_9]] step %[[VAL_3]] iter_args(%[[VAL_16:.*]] = %[[VAL_13]]) -> (f32) { 1265// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_15]]] : memref<?xf32> 1266// CHECK: %[[VAL_18:.*]] = scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_6]] step %[[VAL_3]] iter_args(%[[VAL_20:.*]] = %[[VAL_16]]) -> (f32) { 1267// CHECK: %[[VAL_21:.*]] = scf.for %[[VAL_22:.*]] = %[[VAL_5]] to %[[VAL_7]] step %[[VAL_3]] iter_args(%[[VAL_23:.*]] = %[[VAL_20]]) -> (f32) { 1268// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]], %[[VAL_19]], %[[VAL_22]]] : memref<?x?x?xf32> 1269// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_24]], %[[VAL_17]] : f32 1270// CHECK: %[[VAL_26:.*]] = arith.addf %[[VAL_23]], %[[VAL_25]] : f32 1271// CHECK: scf.yield %[[VAL_26]] : f32 1272// CHECK: } 1273// CHECK: scf.yield %[[VAL_21]] : f32 1274// CHECK: } 1275// CHECK: scf.yield %[[VAL_18]] : f32 1276// CHECK: } 1277// CHECK: memref.store %[[VAL_14]], %[[VAL_12]][] : memref<f32> 1278// CHECK: %[[VAL_30:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<f32> 1279// CHECK: return %[[VAL_30]] : tensor<f32> 1280// CHECK: } 1281func.func @sum_reduction_inv(%arga: tensor<?x?x?xf32>, 1282 %argb: tensor<?xf32, #Td>, 1283 %argx: tensor<f32>) -> tensor<f32> { 1284 %0 = linalg.generic #trait_sum_reduction_inv 1285 ins(%arga, %argb: tensor<?x?x?xf32>, tensor<?xf32, #Td>) 1286 outs(%argx: tensor<f32>) { 1287 ^bb(%a: f32, %b: f32, %x: f32): 1288 %0 = arith.mulf %a, %b : f32 1289 %1 = arith.addf %x, %0 : f32 1290 linalg.yield %1 : f32 1291 } -> tensor<f32> 1292 return %0 : tensor<f32> 1293} 1294 1295#trait_invariants = { 1296 indexing_maps = [ 1297 affine_map<(i,j,k) -> (i)>, // a 1298 affine_map<(i,j,k) -> (j)>, // b 1299 affine_map<(i,j,k) -> (k)>, // c 1300 affine_map<(i,j,k) -> (i,j,k)> // X (out) 1301 ], 1302 iterator_types = ["parallel", "parallel", "parallel"], 1303 doc = "X(i,j,k) = a(i) * b(j) * c(k)" 1304} 1305 1306// CHECK-LABEL: func @invariants( 1307// CHECK-SAME: %[[VAL_0:.*]]: tensor<10xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 1308// CHECK-SAME: %[[VAL_1:.*]]: tensor<20xf32>, 1309// CHECK-SAME: %[[VAL_2:.*]]: tensor<30xf32>, 1310// CHECK-SAME: %[[VAL_3:.*]]: tensor<10x20x30xf32>) -> tensor<10x20x30xf32> { 1311// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 1312// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 10 : index 1313// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 20 : index 1314// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 30 : index 1315// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 1316// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 1317// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10xf32, #sparse_tensor.encoding<{ dimLevelType = [ "dense" ], pointerBitWidth = 0, indexBitWidth = 0 }>> to memref<?xf32> 1318// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : memref<20xf32> 1319// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : memref<30xf32> 1320// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_3]] : memref<10x20x30xf32> 1321// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<10x20x30xf32>) 1322// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 1323// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_14]]] : memref<?xf32> 1324// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 1325// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_16]]] : memref<20xf32> 1326// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_7]] to %[[VAL_6]] step %[[VAL_8]] { 1327// CHECK: %[[VAL_19:.*]] = arith.mulf %[[VAL_15]], %[[VAL_17]] : f32 1328// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_18]]] : memref<30xf32> 1329// CHECK: %[[VAL_21:.*]] = arith.mulf %[[VAL_19]], %[[VAL_20]] : f32 1330// CHECK: memref.store %[[VAL_21]], %[[VAL_13]]{{\[}}%[[VAL_14]], %[[VAL_16]], %[[VAL_18]]] : memref<10x20x30xf32> 1331// CHECK: } 1332// CHECK: } 1333// CHECK: } 1334// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<10x20x30xf32> 1335// CHECK: return %[[VAL_22]] : tensor<10x20x30xf32> 1336// CHECK: } 1337func.func @invariants(%arga: tensor<10xf32, #Td>, 1338 %argb: tensor<20xf32>, 1339 %argc: tensor<30xf32>, 1340 %argx: tensor<10x20x30xf32>) -> tensor<10x20x30xf32> { 1341 %0 = linalg.generic #trait_invariants 1342 ins(%arga, %argb, %argc : tensor<10xf32, #Td>, tensor<20xf32>, tensor<30xf32>) 1343 outs(%argx: tensor<10x20x30xf32>) { 1344 ^bb(%a: f32, %b: f32, %c: f32, %x: f32): 1345 %0 = arith.mulf %a, %b : f32 1346 %1 = arith.mulf %0, %c : f32 1347 linalg.yield %1 : f32 1348 } -> tensor<10x20x30xf32> 1349 return %0 : tensor<10x20x30xf32> 1350} 1351